Antaka: Personalized Tourist Route Recommendation using a Conversational Recommender System Based on Fine-Tuned BERT - Dalam bentuk pengganti sidang - Artikel Jurnal

CLARA ISRA SYAMDAH

Informasi Umum

Kode

25.04.7016

Klasifikasi

000 - General Works

Jenis

Karya Ilmiah - Skripsi (S1) - Reference

Subjek

Recommender Systems

Dilihat

36 kali

Informasi Lainnya

Abstraksi

This paper introduces Antaka, a Conversational Recommender System (CRS) designed for personalized tourist route planning in Indonesia. Unlike traditional recommender systems that rely on content-based or collaborative filtering methods, An taka incorporates natural language understanding through the BERT Transformer model to dynamically capture user preferences during multi-turn conversations. The system is fine-tuned using a Named Entity Recognition (NER) approach on Indonesian tourism datasets to identify key entities, including destination categories, cities, price ranges, and user rating preferences. Antaka’s architecture enables real-time interaction and personalized suggestions tailored to the user’s intent and context. Experimental results show significant improvements after fine-tuning: precision increased from 0.53 to 0.90, recall from 0.09 to 0.86, F1-score from 0.06 to 0.88, and accuracy reached 0.86. In recommendation evaluations, Antaka achieved a Mean Reciprocal Rank (MRR) of 0.87 and Precision@3 of 0.79. These findings highlight Antaka’s potential to deliver relevant and contextual route recommendations through natural conversations, offering a novel contribution to CRS research in low-resource languages and explicitly addressing the Indonesian tourism landscape.